1
The arch not the stones: Universal feature theory without universal features
B. Elan Dresher (University of Toronto)
In Invisible Cities, Italo Calvino (1974) imagines an exchange between
Marco Polo and Kublai Khan. Marco Polo describes a bridge, stone by
stone.
“But which is the stone that supports the bridge?” Kublai Khan asks.
“The bridge is not supported by one stone or another,” Marco
answers, “but by the line of the arch that they form.”
Kublai Khan remains silent, reflecting. Then he adds: “Why do you
speak to me of the stones? It is only the arch that matters to me.”
Polo answers: “Without stones there is no arch.”
1.
Introduction
I will propose that the phonological component of the grammar computes
features, but these features are not innate. Rather, they are created by the learner
as part of the acquisition of phonology. Further, Universal Grammar (UG)
requires that these features be organized into contrastive feature hierarchies that
reflect phonological activity and the contrasts in the lexical inventory.
The paper is organized as follows: In section 2, I review the objections
that have been raised against innate distinctive features, and I introduce the notion
of contrastive feature hierarchies. In section 3, I present an extended example,
2
based on the Classical Manchu vowel system, illustrating the relation between
contrast and phonological activity. In section 4, I consider the implications of this
approach for the acquisition of phonological representations. Section 5 takes up
the relationship between contrastive hierarchies and phonetic ‘substance’, with
examples drawn from the tonal organization in Chinese dialects. Section 6
concludes with a brief discussion of the place of phonology in the Faculty of
Language.
2.
Phonological features: What is universal?
There is a growing consensus that phonological features are not innate, but rather
‘emerge’ in the course of acquisition. In a recent volume titled Where do
phonological features come from? (Clements & Ridouane 2011), most of the
papers take an emergentist position; none argue for innate features. Mielke (2008)
and Samuels (2011) summarize the arguments against innate features (1).
(1)
Arguments against innate features
a. From a biolinguistic perspective, phonological features are too
specific, and exclude sign languages (van der Hulst 1993;
Sandler 1993).
b. Empirically, no one set of features have been discovered that
‘do all tricks’ (Hyman 2010 with respect to tone features, but
the remark applies more generally).
3
c. Since at least some features have to be acquired from
phonological activity, a prespecified list of features becomes
less useful in learning.
But if features are emergent, we need to explain why they are required at
all, and what UG principles account for the way they function in the phonology. I
propose that the task of the learner is to arrive at a set of features that account for
the contrasts and the phonological activity in a given language.
To implement contrast in an explicit theory, I borrow an idea from
Jakobson and his collaborators (Jakobson, Fant & Halle 1952; Jakobson & Halle
1956), that was called ‘branching trees’ in the literature of the 1950s and 1960s
(2); I call it the Successive Division Algorithm (Dresher 1998b; 2003; 2009).
(2)
Branching feature trees
Contrastive features are assigned by language-particular feature
hierarchies.
(3)
The Successive Division Algorithm
Assign contrastive features by successively dividing the inventory
until every phoneme has been distinguished.
As a first approximation, I assume further that phonology computes only
contrastive features, in keeping with the Contrastivist Hypothesis, which Hall
(2007) formulates as in (4).
4
(4)
The Contrastivist Hypothesis (Hall 2007)
The phonological component of a language L operates only on
those features which are necessary to distinguish the phonemes of
L from one another.
According to (4), only contrastive features can be phonologically active. If
this hypothesis is correct, it follows as a corollary that if a feature is
phonologically active, then it must be contrastive (5).
(5)
Corollary to the Contrastivist Hypothesis
If a feature is phonologically active, then it must be contrastive.
One final assumption is that features are binary, and that every feature has
a marked and unmarked value. I assume that markedness is language particular
(Rice 2003; 2007) and accounts for asymmetries between the two values of a
feature, where these exist. I will designate the marked value of a feature F as
‘[F]’, and the unmarked value as ‘(non-F)’.
(6)
Assumption about phonological primes
Features are binary; every feature has a marked value, designated
‘[F]’, and an unmarked value designated ‘(non-F)’.
It is a working assumption that the phonological primes are binary
features. It is an empirical hypothesis that the learner creates binary features and
not other sorts of entities, such as privative elements or dependency structures of
various kinds. As long as these other types of representations are compatible with
5
the contrastive hierarchy and related assumptions as outlined above, we can
consider them to be variations of the theory presented here.
For the content of features (or whatever primes are assumed), I assume
that learners make use of the available materials relevant to the modality: for
spoken language, acoustic and articulatory properties of speech sounds; and for
sign language, hand shapes and facial expressions.
On this view, the concept of a contrastive hierarchy is an innate part of
UG, and is the glue that binds phonological representations and makes them
appear similar from language to language.
For example, if a language has three vowel phonemes /i, a, u/, and if the
vowels are split off from the rest of the inventory so that they form a subinventory, then they must be assigned a contrastive hierarchy with two vowel
features. Though the features and their ordering vary, the limit of two features
constrains what the hierarchies can be. Two possible contrastive hierarchies using
the features [back] and [low] are shown in (7).
(7)
Contrastive hierarchies for /i, a, u/ with [back] and [low]
a. [back] > [low]
[syllabic]
ei
[back]
(non-back)
ty
g
[low] (non-low) /i/
g
g
/a/
/u/
b. [low] > [back]
[syllabic]
ru
[low]
(non-low)
g
ty
/a/ [back] (non-back)
g
g
/u/
/i/
Here are two more hierarchies, using [high] and [round].
6
(8)
Contrastive hierarchies for /i, a, u/ with [high] and [round]
a. [high] > [round]
[syllabic]
ru
[high]
(non-high)
ty
g
[rnd] (non-rnd) /a/
g
g
/u/
/i/
b. [round] > [high]
[syllabic]
ru
[round] (non-round)
g
ty
/u/
[hi] (non-hi)
g
g
/i/
/a/
The contrastive hierarchies in (7) and (8) constrain phonological activity.
Thus, if only contrastive features can be phonologically active, it follows that both
/a/ and /u/ can potentially trigger backing in (7a), only /u/ is a potential backing
trigger in (7b), and no vowel is expected to trigger backing in (8). By the same
token, the languages in (7) could have a lowering rule triggered by /a/, whereas
the languages in (8) might have raising, triggered by both /u/ and /i/ in (8a), and
by only /i/ in (8b). Spahr (2012) proposes that the intermediate nodes in
contrastive trees can also be referred to by the phonology in ‘archiphonemic’
fashion. Thus, we might find that /a/ and /u/ in (7a) merge in vowel reduction,
suspending the [low] contrast and deriving a vowel that is contrastively [back];
whereas in (8b) vowel reduction might merge /i/ and /a/.
If the contrastive hierarchy governs phonological organization, we might
also expect it to influence phonological change, as has been proposed by Harvey
(2012), Ko (2012), and Oxford (2012). In particular, we might expect the
hierarchy to constrain neutralization and merger, on the assumption that mergers
7
affect phonemes that are contrastive sisters. Thus, in (7a) we might expect /u/ to
merge with /a/, whereas in (7b) and (8a) it is more likely to merge with /i/.
Typological generalizations can thus not be found by looking at
inventories alone (say, /i, a, u/), or at individual phonemes (say, /a/), or phones
([a]), without also considering the relevant contrastive feature hierarchy.
It is important to emphasize that, though phonological features may make
use of innate auditory dispositions, they are not the same as those, but are
cognitive entities created by learners. Thus, the contrasts indicated by [back] and
[low] may be cross-linguistically common because we have neurons sensitive to
formant transitions. So, it appears, do ferrets (Mesgarani et al. 2008). But ferrets
do not necessarily have our kind of phonological representations.
Notice also that, on this view, lexical specifications are limited to
contrastive features, so are not pronounceable. In (7a), for example, the phoneme
designated /u/ has only two features: [back] and (non-low). Why, then, is it
designated /u/ and not /ʊ/, /ʌ/, /ɯ/, /ɨ/ or /o/, among other choices? As far as its
contrastive status goes, any of these alternatives would be equally appropriate.
We could indicate the phonemes as in (9), for example, though these symbols are
typographically less convenient.
8
(9)
Underspecification of contrastive hierarchies for /i, a, u/
a. [back] > [low]
[syllabic]
ei
[back]
(non-back)
ty
g
[low] (non-low) /œ/
g
g
/ɒ/
/ɯ/
b. [low] > [back]
[syllabic]
ru
[low]
(non-low)
g
ty
/æ/ [back] (non-back)
g
g
/ɨ/
/e/
Unless the vowels are further specified in the phonology by other
contrastive features (originating in the consonants, for example), they are made
more specific only in a post-phonological component. Stevens, Keyser &
Kawasaki (1986) proposed that feature contrasts can be enhanced by other
features that have similar acoustic effects. Hall (2011) shows how the
enhancement of contrastive features can result in configurations predicted by
Dispersion Theory (Liljencrants & Lindblom 1972; Lindblom 1986; Flemming
2002). Thus, a non-low back vowel can enhance these features by being round
and high, that is, [u]. These enhancements are not necessary, however, and other
realizations are possible.
It is thus the contrastive hierarchy, not the features, that is native to UG,
the thing that ‘matters’ to us (Kublai Khan, op. cit.). In the words of Jakobson,
Fant & Halle (1952: 9): “The dichotomous scale is the pivotal principle of the
linguistic structure. The code imposes it upon the sound”. So why should we
concern ourselves with features at all? Because without features there is no
hierarchy.
9
3.
The connection between contrast and activity
In this section I will illustrate the connection between contrast and phonological
activity, taking as an extended example the Classical Manchu vowel system,
following the analysis of Zhang (1996) and Dresher & Zhang (2005). Classical
Manchu has six vowel phonemes, as shown in (10).
(10)
Classical Manchu vowel system (Zhang 1996)
/i/
/u/
/ʊ/
/əә/
/ɔ/
/a/
Even if there were innate universal features, there would be considerable
ambiguity as to how they apply to this system. For example, where is the
boundary between the low vowel(s) and the non-low vowels? How many heights
should we distinguish: two, three, five? For further insight, we need to look at
how the vowels pattern, that is, at the types of phonological activity they exhibit.
The three most notable kinds of phonological activity involving vowels
are ATR harmony, labial (rounding) harmony, and palatalization. I will briefly
discuss each in turn.
3.1
ATR harmony
The vowels /əә/ and /u/ trigger ATR harmony within a word: /əә/ alternates with /a/
and /u/ alternates with /ʊ/.
10
(11)
ATR harmony
a. /əә/ alternates with /a/
[ATR]
xəәxəә
(non-ATR) aɢa
‘woman’
xəәxəә-ŋgəә
‘female’
‘rain’
aɢa-ŋɢa
‘of rain’
xəәrəә-ku
‘ladle’
b. /u/ alternates with /ʊ/
[ATR]
xəәrəә- ‘ladle out’
(non-ATR) paqtʼa- ‘contain’
paqtʼa-qʊ ‘internal organs’
An apparent exception is caused by the fact that /ʊ/ changes to [u]
everywhere except after dorsal (velar ~ uvular) consonants; however, the
underlying contrast between /ʊ/ and /u/ emerges in the way they participate in
ATR harmony (12).
(12)
ATR harmony and the neutralization of /u/ and /ʊ/
a. Underlying /u/: ATR harmony
[ATR]
susəә
‘coarse’
[ATR]
xəәtʼu ‘stocky’
susəә-təә-
‘make coarsely’
xəәtʼu-kəәn ‘somewhat stocky’
b. Underlying /ʊ/: non-ATR vowels
(non-ATR) tulpa
‘careless’
(non-ATR) tatʼʂun ‘sharp’
tulpa-ta- ‘act carelessly’
tatʼʂu-qan ‘somewhat sharp’
The vowel /i/ is neutral and co-occurs with both ATR and non-ATR
vowels (13).
11
(13)
/i/ is neutral with respect to ATR harmony
a. /əә/ ~ /a/ suffix
[ATR]
pəәki
‘firm’
(non-ATR) paqtʂʼin ‘opponent’
pəәki-ləә
‘make firm’
paqtʂʼi-la- ‘oppose’
b. /u/ ~ /ʊ/ suffix
[ATR]
sitəәrəә- ‘hobble’
(non-ATR) panjin ‘appearance’
sitəәrəә-sxun ‘hobbled/lame’
panji-sχʊn ‘having money
c. /i/ suffix
[ATR]
əәmtʼəә
(non-ATR) taχa-
‘one each’
əәmtʼəә-li
‘alone; sole’
‘follow’
taχa-li
‘the second’
When /i/ is in a position to trigger harmony, it occurs only with non-ATR
vowels (14).
(14)
Stems with only /i/ co-occur with non-ATR vowels
a. /əә/ ~ /a/ suffix
(non-ATR) ili-
‘stand’
ili-χa
‘stood’
(non-ATR) fili
‘solid’
fili-qan ‘somewhat solid’
b. /u/ ~ /ʊ/ suffix
(non-ATR) tʂʼili- ‘to choke’
tʂʼili-qʊ ‘choking’
(non-ATR) sifi-
sifi-qʊ
‘stick in the hair’
‘hairpin’
The evidence from activity, therefore, is that /əә/ and /u/ have an active
feature in common that is not shared by the other vowels; by hypothesis, this
feature must be contrastive. What feature could this be? I have already given
12
away that it is [ATR]. But this is not obvious right away, because though /əә/ and
/u/ are phonetically ATR (= {ATR}), so is /i/. But there is no obvious alternative,
so we can designate the feature as [ATR]. The learner will have to find a feature
ordering in which the feature applies to /əә/ and /u/, but not /i/.
3.2
Labial (rounding) harmony
Two successive /ɔ/ vowels cause a suffix /a/ to become /ɔ/ (15a); a single /ɔ/, short
or long, does not trigger rounding (15b).
(15)
Labial (rounding) harmony
a. Two successive /ɔ/ vowels trigger labial harmony
ɔ…ɔ
pɔtʂʼɔ ‘colour’
pɔtʂʼɔ-ŋɢɔ ‘coloured’
Compare
aɢa
aɢa-ŋɢa
‘rain’
‘of rain’
b. A single /ɔ/, short or long, does not trigger rounding
Single ɔ
tɔ-
‘alight (birds)’
Single ɔɔ
tɔɔ- ‘cross (river)’
tɔ-na-
‘alight in swarm’
tɔɔ-na- ‘go to cross’
Note that /u/ and /ʊ/ do not trigger labial harmony (16).
(16)
No labial harmony triggered by high vowels
a. After /u/
gulu
‘plain’
gulu-kəәn
‘somewhat plain’
kumun
‘music’
kumu-ŋgəә ‘noisy’
b. After /ʊ/ (/ʊ/ becomes [u] except after a back consonant)
χʊtun
‘fast’
χʊtu-qan
‘somewhat fast’
tursun
‘form’
tursu-ŋɢa
‘having form’
13
The evidence from activity here, then, is that /ɔ/ must have an active,
therefore contrastive, feature that causes rounding. [labial] is an obvious
candidate. But /u/ and /ʊ/ are also phonetically {labial}, though there is no
evidence that that they have an active [labial] feature. Here, the preferred analysis
is one where contrastive [labial] is restricted to /ɔ/, and excludes /u/ and /ʊ/.
3.3
Palatalization
The vowel /i/ uniquely causes palatalization of a preceding consonant, which
suggests that it alone has a contrastive triggering feature we call [coronal]. In this
case /i/ is the only vowel that falls in the space of the phonetic percept {coronal}.
3.4
Height contrast
The alternations /əә/ ~ /a/ ~ /ɔ/ and /u/ ~ /ʊ/ are limited to a height class, and we
still need to distinguish /əә/ from /u/ and /a/ from /ʊ/. It is simplest to assume one
height contrast, which we call [low]. Since height is a relative property, it is not a
problem to base the contrastive feature on a perceptible phonetic difference based
on relative height or sonority. As there are only two height classes, [high] would
also be possible here.
3.5
Contrastive feature hierarchy for Classical Manchu
Putting together the evidence of phonological activity surveyed to here, we need
to arrive at a feature hierarchy that yields the values identified above. Zhang
(1996) proposes the hierarchy: [low] > [coronal] > [labial] > [ATR], illustrated by
the tree in (17), and yielding the representations in (18).
14
(17)
Contrastive hierarchy for Classical Manchu vowels
[syllabic]
qp
(non-low)
ei
[coronal]
(non-coronal)
g
ru
/i/
[ATR]
(non-ATR)
g
g
/u/
/ʊ/
(18)
[low]
ei
(non-labial)
[labial]
ru
g
[ATR]
(non-ATR)
/ɔ/
g
g
/əә/
/a/
Classical Manchu vowels: Contrastive featural representations
/i/
/u/
[coronal]
[ATR]
/ʊ/
/a/
[low]
/əә/
[low]
[ATR]
/ɔ/
[low]
[labial]
Though /i/ is phonetically {ATR}, it lacks a contrastive [ATR] feature, so
does not participate in ATR harmony. Similarly, /u/ and /ʊ/ are phonetically
{labial}, but lack a contrastive [labial] feature, so they do not participate in
rounding harmony.
4.
The contrastive hierarchy and the acquisition of phonological
representations
The phonological contrastive features that must be acquired are not identical to
the acoustic percepts that can be detected in the signal by an early learner. We
have already seen that the phonetic ranges of the acoustic percepts on which the
Manchu features are based are not coextensive with the phonological
representations of the vowels. Percepts are not representations. The early learner
15
can perceive and distinguish speech sounds, but has not yet identified any
contrasts in the vowel system.
Assume that contrastive features are determined in order. Assuming that
vowels are differentiated early from consonants, at the outset the learner posits
one undifferentiated vowel phoneme /V/. At some point the learner discovers a
contrast between a [low] vowel /A/ and a non-low vowel /I/ (the symbols are for
convenience only). The non-low vowels split into a [coronal] vowel /i/ and a noncoronal vowel /U/. The low vowels split into a [labial] vowel /ɔ/ and a non-labial
vowel /A/. [labial] has nothing to do in the non-low vowels, where [coronal] has
arrived first. Finally, [ATR] makes the final set of contrasts. [ATR] cannot apply
to the [coronal] vowel which is already uniquely specified. Nor can it apply to the
[labial] vowel, which is not even phonetically {ATR}. At this point there are no
further contrasts to find in this language, and the hierarchy is complete.
The preceding is a hypothetical sequence showing the order that contrasts
in Manchu vowels would be acquired, if the order of acquisition mirrors the
feature hierarchy of the adult system (this assumption may not be correct: we may
have to allow for false starts and revisions, particularly at early stages). This idea
stems from Jakobson (1941) and is a natural way to describe developing
phonological inventories (Pye, Ingram and List 1987; Ingram 1988; 1989; Levelt
1989; Dinnsen et al. 1990; Dinnsen 1992; 1996; see Dresher 1998a for a review).
In the formulation of Rice and Avery (1995), phonological representations are
built into systems of increasing complexity, based on the input from phonetic
16
perception together with evidence from the grammar, which itself becomes more
complex and removed from the initial percepts.
Fikkert (1994) presents observed acquisition sequences in the
development of Dutch onsets that follow this general scheme. At Stage 1 in (19),
there are no contrasts. The value of the consonant defaults to the least marked
onset, namely an obstruent plosive. At Stage 2, the first contrast is made, between
obstruents and sonorants. The former remains the unmarked option (designated as
u. The marked option (m), sonorant, defaults to nasal. Following this stage,
children differ. Some expand the obstruent branch first, bringing in marked
fricatives in contrast with plosives (Stage 3a). Others expand the sonorant branch,
introducing marked sonorants, either liquids or glides (Stage 3b). Continuing in
this way we will eventually have a tree that gives all and only the contrasting
features in the language.
(19)
Development of Dutch onset consonants
consonant
uqpm
obstruent
sonorant
urum
urum
plosive (3a) fricative
nasal (3b) liquid/glide
g
g
g
g
/P/
/F/
/N/
/L/J/
Stage 1
Stage 2
Stages 3a, 3b
The scenario presented here has been criticized by Hale & Reiss (2008)
and Samuels (2009). Hale & Reiss (2008: 39–42) argue that the “traditional
model” set out by Rice & Avery (1995), and adopted here, cannot be accepted:
17
“We believe that such a learning path is not possible given standard assumptions
about the language faculty…Rice and Avery’s theory, and those like it, must be
rejected as unparsimonious, incompatible with the generative program, and
incapable of modeling a successful learning path.”
To illustrate their argument, imagine a five-vowel language /i, e, a, o, u/,
and suppose that a learner has reached a stage that distinguishes only three vowels
with the features shown in (20).
(20)
Acquiring a five-vowel language, only three vowels distinguished
(non-back)
[back]
/i/
/u/
/I/
/U/
/e/
/o/
/a/
(non-low)
[low]
The learner has not yet made a distinction between /i/ ~ /e/ and /u/ ~ /o/;
that is, the feature [high] has not yet been acquired. Thus, words like [pit] and
[pet] are both represented /pIt/; for the child, the vowels in both have the identical
representation (non-low, non-back). How do learners get to the five-vowel stage?
Hale & Reiss (2008) argue that they can’t get there from here!
“[A]ny vowel that the child is presented with must be parsed as one of the
three, or else it will not be parsed at all... A representation can only be assigned
using the available representational apparatus.” Thus, a learner that can
distinguish [i] from [e] must already have access to the feature [high], contrary to
what is assumed here.
18
Our answer is that learners can discriminate sounds for which they have
no phonological representations using their innate perceptual system, but Hale &
Reiss do not accept this explanation, either. They argue that the grammar should
not make use of two types of representation, “a phonological representation,
which starts out with access to a minimal set of features, and a phonetic or
acoustic representation, which makes use of fully specified phonetic feature
matrices.” But we are not proposing that what I have called ‘acoustic percepts’ are
part of the grammar, or make use of feature matrices. Hale & Reiss, however,
reject the notion that learners can make use of “raw acoustic images” that are
outside the grammar. They assert that any discrimination of speech sounds must
be in terms of innate phonological features. But recall the evidence that ferrets
have neurons that can discriminate fine details of speech sounds. Presumably we
have them, too, before and after they enable us to acquire phonological
representations.
Hale & Reiss’s view is that phonological representations begin as very
detailed and become simpler in the course of acquisition. This theory requires that
phonological features are innate, universal, and unambiguous, an untenable
position. In support of their view, they cite evidence that infants begin by
attending to many potential sources of contrasts, and are more able than adults to
discriminate sounds not used in the ambient language (Eimas et al. 1971; Werker
et al. 1981). That is, acquisition of the native language requires that they ‘tune’
their perceptual system to the contrasts used in their language, while learning to
19
disregard contrasts that are not used (Werker & Tees 1984). However, there is no
evidence that this ‘tuning’ applies to phonological representations. The
observations about infants apply to phones, not to phonemes. Learning to ignore
sounds and distinctions that are not relevant to their native language is obviously
helpful in eventually acquiring phonological features, but it is not the same
process. In fact, if we combine the studies of infants tuning phonetic perceptions
with phonological studies of the role of contrast in phonological inventories, we
obtain a picture of a learner going in two contrary directions simultaneously: the
perceptual system is learning to ignore irrelevant contrasts, while phonological
representations are becoming more complex.
In sum, the learning path proposed by Hale & Reiss is untenable because it
requires an innate and universal set of features, and it further assumes that
learners can immediately assign the correct featural representations to any surface
sound they hear (that is, features must also be unambiguous). None of these
prerequisites are met by phonological features, given the problems with assuming
innate features, and the ambiguity of many features: How low qualifies as [low]?
What height differences can be tolerated in segments of the same height? Where
is the boundary between [back] and (non-back)?
It appears that there is no alternative but to suppose that learners must be
able to perceive distinctions that are not yet encoded in their grammar. It follows
that perception is not limited by the current grammar. For that matter, adults are
able to perceive unfamiliar phonetic distinctions in a foreign language if they
20
focus on them. I conclude that Hale & Reiss’s (2008) arguments against Rice &
Avery (1995), and by implication the account presented here, do not go through,
and that the alternative acquisition path they propose is not tenable. Thus there is
no obstacle to adopting the ‘traditional’ view that phonological representations
become more complex in the course of acquisition, and that learners acquire the
contrasts of their language in stages. The contrastive hierarchy provides a way of
connecting accounts in the acquisition literature of developing inventories with
synchronic and diachronic phonology.
5.
The contrastive hierarchy and phonetic ‘substance’
Krekoski (2013) constructs contrastive trees for the tone systems of a number of
languages that descend from Middle Chinese. He bases the trees not on the
phonetics of the tones, but on the patterns of activity they display in the form of
tone sandhi. Thus, Beijing Mandarin has the four tones in (21), which participate
in two robust sandhi rules (22).
(21)
(22)
Beijing Mandarin tones
a. /55/
high level
c. /214/
low concave
b. /35/
rising
d. /51/
high falling
Beijing Mandarin tone sandhi
a. /214/ → 35/_____/214/
b. /35/
→ 55/{/35/, /55/}_____T where (T = any tone)
21
Krekoski (2013) assumes that, where possible, tones related by a sandhi
rule differ minimally, that is by only one feature. Thus, tone /35/ differs by one
feature from /214/ and from /55/. A tree satisfying these constraints is given in
(23): [α] and [β] are placeholders for features which will be given a phonetic
interpretation.
(23)
Beijing Mandarin contrastive hierarchy
T
wo
[+α]
[−α]
ty
ty
[+β]
[−β]
[+β] [−β]
g
g
g
g
/55/
/35/
/51/ /214/
Pingyao is a Jin language with four underlying tones (24). Though two of
them have merged at the surface, they can be distinguished by the way they
participate in tonal alternations. Krekoski identifies nine tone sandhi rules in
Pingyao. Their inputs and outputs are summarized in (25); I omit alternations that
are purely allotonic.
(24)
Pingyao tones
a. /13a/ low rising
c. /53/ high falling
b. /13b/ low rising
d. /35/ high rising
22
(25)
Pingyao tone sandhi
Input
Outputs
a. /13a/
35
b. /35/
13 [= 13a], 53
c. /53/
35, 13 [= 13b]
Though the two [13] tones sound the same, Krekoski gives arguments for
associating them with either /13a/ or /13b/. Following the same procedure as for
Beijing, Krekoski arrives at a tree for Pingyao whereby each of the tonal
alternations involves a change of only one feature.
(26)
Pingyao (Jin) contrastive hierarchy
T
wo
[+α]
[−α]
ty
ty
[+β]
[−β]
[+β]
[−β]
g
g
g
g
/13a/
/13b/
/35/
/53/
Krekoski observes that Beijing and Pingyao tones in corresponding
positions in the trees in (23) and (26) are cognates, and descend from the same
Middle Chinese tone. That is, despite extensive changes in their phonetics, the
tones retain the same positions in the contrastive hierarchy.
Up to here we have not tried to give the features phonetic interpretations;
however, features are not purely abstract entities. Krekoski (2013) suggests
correlates for the features as in (27) and (28) (I do not attempt to assign
23
markedness to them). The feature [extreme] in (28) refers to the periphery of a
tonal space, and [inner] designates a more central region of the space.
(27)
Beijing Mandarin tone features
T
qp
[non-falling]
[falling]
ru
ru
[high] [non-high] [high] [non-high]
g
g
g
g
/55/
/35/
/51/
/214/
(28)
Pingyao (Jin) tone features
T
qp
[low]
[high]
ru
ru
[inner] [extreme]
[inner] [extreme]
g
g
g
g
/13a/
/13b/
/35/
/53/
Following the same methodology, Krekoski posits the tree in (29) for
Tianjin Mandarin. Surprisingly, these tones do not correspond as expected with
their cognates in Beijing and Pingyao. Tones /21/ and /53/ are in the ‘wrong
place’ relative to the other dialects that descend from Middle Chinese.
(29)
Tianjin Mandarin contrastive hierarchy
T
wo
[+α]
[−α]
ty
ty
[+β]
[−β]
[+β]
[−β]
g
g
g
g
/53/
/45/
/21/
/213/
24
Tracing the tones from Middle Chinese, Krekoski proposes that an earlier
stage of Tianjin (*Proto-Tianjin) must have had the hierarchy in (30).
(30)
*Proto-Tianjin Mandarin contrastive hierarchy
T
wo
[+α]
[−α]
ty
ty
[+β]
[−β]
[+β]
[−β]
g
g
g
g
/21/
/45/
/53/ /213/
Why did a contrastive shift occur in the history of Tianjin? An answer can
be found in the phonetics of the tones. Krekoski observes that it is difficult to find
plausible phonetic correlates for the features in *Proto-Tianjin, whereas the
current Tianjin system clearly groups the tones by height. He proposes: “Tonal
drift likely accreted changes in height values until the system of contrasts reached
some critical inflection point which precipitated the reanalysis of specifications.”
What this example illustrates is that features may be suggested by patterns
of phonological activity, but that phonetic substance also has a say. Contrastive
trees for tonal features can remain stable even as the phonetic realizations of the
tones change; but the feature tree is restructured when it gets too out of sync with
the phonetics. Without such a mechanism, we would expect a much greater
proliferation of ‘crazy rules’ than we actually find.
While phonetic substance influences the contrastive feature hierarchy, the
influence is not all in this direction. I argued above that the contrastive hierarchy
serves as an organizing principle for synchronic phonology, and influences the
25
direction of diachronic changes, such as mergers. The conclusion is that influence
runs in both directions.
6.
Conclusion: Phonology and the Faculty of Language
The approach presented here shares with ‘substance-free’ theories the idea that
the phonetic content of features is not part of the phonological computational
system (Blaho 2008; Hale & Reiss 2000a; 2000b; 2008; Iosad 2012; Morén
2003; 2006; 2007; Samuels 2009; 2011). Some of these theories go too far, in my
view, in shifting the explanation for phonological patterning to external factors. In
his review of Samuels (2011), Hall (2012: 738) comments: “the substance-free
and the substance-based views are alike in that they both posit functional phonetic
explanations for substantive phonological patterns…the two lines of thought, in
their different ways, both turn away from the practice of constructing formal
explanations for substantive patterns.” The contrastive feature hierarchy restores
the balance between functional and formal explanations, to the extent that it
serves as a formal organizing principle of the phonology.
More generally, it has been suggested that only syntactic recursion is part
of the narrow faculty of language (FLN; Hauser, Chomsky & Fitch 2002), and
that phonology is outside FLN. However, the contrastive hierarchy has a
recursive digital character, like other aspects of FLN. Like syntax, phonology
takes substance from outside FLN and converts it to objects that can be
manipulated by the linguistic computational system. The parallels between
26
phonology and syntax may go even further, if it turns out that syntax, too, is in the
business of creating contrastive hierarchies of morphosyntactic features (Cowper
& Hall 2013).
Acknowledgments
For discussions, ideas, and analyses I would like to thank Elizabeth Cowper,
Daniel Currie Hall, Paula Fikkert, Ross Godfrey, Christopher Harvey, Ross
Krekoski, Will Oxford, Keren Rice, Christopher Spahr, and Zhang Xi, and other
members of the project on Markedness and the Contrastive Hierarchy in
Phonology at the University of Toronto (Dresher and Rice 2007).
References
Blaho, Sylvia. 2008. The syntax of phonology: A radically substance-free
approach. Doctoral dissertation, University of Tromsø.
Calvino, Italo. 1974. Invisible cities. Translated from the Italian by William
Weaver. New York: Harcourt Brace Jovanovich.
Clements, G. Nick & Rachid Ridouane (eds.). 2011. Where do features come
from? Cognitive, physical and developmental bases of distinctive speech
categories. Amsterdam: John Benjamins.
Cowper, Elizabeth & Daniel Currie Hall. 2013. Reductiō ad discrīmen: Where
features come from. Paper presented at Conference on Features in
27
Phonology, Morphology, Syntax and Semantics: What are they? CASTL,
University of Tromsø, October–November 2013.
Dinnsen, Daniel A. 1992. Variation in developing and fully developed phonetic
inventories. In Charles A. Ferguson, Lisa Menn & Carol Stoel-Gammon
(eds.), Phonological development: Models, research, implications, 191–
210. Timonium, MD: York Press.
Dinnsen, Daniel A. 1996. Context-sensitive underspecification and the acquisition
of phonetic contrasts. Journal of Child Language 23. 31–55.
Dinnsen, Daniel A., Steven B. Chin, Mary Elbert & Thomas W. Powell. 1990.
Some constraints on functionally disordered phonologies: Phonetic
inventories and phonotactics. Journal of Speech and Hearing Research 33.
28–37.
Dresher, B. Elan. 1998a. Child phonology, learnability, and phonological theory.
In Tej Bhatia & William C. Ritchie (eds.), Handbook of language
acquisition, 299–346. New York: Academic Press.
Dresher, B. Elan. 1998b. On contrast and redundancy. Paper presented at the
annual meeting of the Canadian Linguistic Association, Ottawa. Ms.,
University of Toronto.
Dresher, B. Elan. 2003. Contrast and asymmetries in inventories. In Anna-Maria
di Sciullo (ed.), Asymmetry in grammar, volume 2: Morphology,
phonology, acquisition, 239–57. Amsterdam: John Benjamins.
28
Dresher, B. Elan. 2009. The contrastive hierarchy in phonology. Cambridge:
CUP.
Dresher, B. Elan and Keren Rice. 2007. Markedness and the contrastive hierarchy
in phonology. http://homes.chass. utoronto.ca/~contrast/.
Dresher, B. Elan and Xi Zhang. 2005. Contrast and phonological activity in
Manchu vowel systems. Canadian Journal of Linguistics/Revue
canadienne de linguistique 50: 45–82.
Eimas, Peter D., Einar R. Siqueland, Peter Jusczyk and James Vigorito. 1971.
Speech perception in infants. Science 171. 303–306.
Fikkert, Paula. 1994. On the acquisition of prosodic structure (HIL Dissertations
6). Dordrecht: ICG Printing.
Flemming, Edward S. 2002. Auditory representations in phonology. London:
Routledge.
Hale, Mark and Charles Reiss. 2000a. “Substance abuse” and “dysfunctionalism”:
Current trends in phonology. Linguistic Inquiry 31. 157–169.
Hale, Mark and Charles Reiss. 2000b. Phonological theory without substance. In
Noel Burton-Roberts, Philip Carr & Gerard Docherty (eds.), Phonological
knowledge: Conceptual and empirical foundations, 161–184. Oxford:
OUP.
Hale, Mark & Charles Reiss. 2008. The phonological enterprise. Oxford: OUP.
Hall, Daniel Currie. 2007. The role and representation of contrast in phonological
theory. Doctoral dissertation, University of Toronto.
29
Hall, Daniel Currie. 2011. Phonological contrast and its phonetic enhancement:
Dispersedness without dispersion. Phonology 28: 1–54.
Hall, Daniel Currie. 2012. Review of Bridget D. Samuels, Phonological
architecture: A biolinguistic perspective. Journal of Linguistics 48. 736–
741.
Harvey, Christopher. 2012. Contrastive shift in Ob-Ugric vowel systems. Ms.,
Department of Linguistics, University of Toronto.
Hauser, Marc D., Noam Chomsky and W. Tecumseh Fitch. 2002. The faculty of
language: What is it, who has it, and how did it evolve? Science 298.
1569–1579.
Hulst, Harry van der. 1993. Units in the analysis of signs. Phonology 10. 209–
241.
Hyman, Larry M. 2011. Do tones have features? In John A. Goldsmith, Elizabeth
Hume & W. Leo Wetzels (eds.), Tones and features, 50–80. Berlin:
Mouton De Gruyter.
Ingram, David. 1988. Jakobson revisited: Some evidence from the acquisition of
Polish phonology. Lingua 75. 55–82.
Ingram, David. 1989. First language acquisition: Method, description and
explanation. Cambridge: CUP.
Iosad, Pavel. 2012. Representation and variation in substance-free phonology: A
case study in Celtic. Doctoral dissertation, University of Tromsø.
30
Jakobson, Roman. 1941. Kindersprache, Aphasie, und allgemeine Lautgesetze.
Uppsala: Uppsala Universitets Arsskrift. Translated by A. R. Keiler as
Child language, aphasia, and phonological universals. The Hague:
Mouton, 1968.
Jakobson, Roman, C. Gunnar M. Fant and Morris Halle. 1952. Preliminaries to
Speech Analysis. MIT Acoustics Laboratory, Technical Report, No. 13.
Reissued by MIT Press, Cambridge, Mass., Eleventh Printing, 1976.
Jakobson, Roman and Morris Halle. 1956. Fundamentals of language. The
Hague: Mouton.
Ko, Seongyeon. 2012. Tongue root harmony and vowel contrast in Northeast
Asian languages. Doctoral dissertation, Cornell University.
Krekoski, Ross. 2013. On tone and the nature of features. Ms., Department of
Linguistics, University of Toronto.
Levelt, Clara C. 1989. An essay on child phonology. M.A. thesis, Leiden
University.
Liljencrants, Johan & Björn Lindblom. 1972. Numerical simulation of vowel
quality systems: The role of perceptual contrast. Language 48. 839–862.
Lindblom, Björn. 1986. Phonetic universals in vowel systems. In John J. Ohala
and Jeri J. Jaeger (eds.), Experimental phonology, 13–44. New York:
Academic Press.
31
Mesgarani, Nima, Stephen V. David, Jonathan B. Fritz & Shihab A. Shamma.
2008. Phoneme representation and classification in primary auditory
cortex. Journal of the Acoustical Society of America 123. 899–909.
Mielke, Jeff. 2008. The emergence of distinctive features. Oxford: OUP.
Morén, Bruce. 2003. The Parallel Structures Model of feature geometry. Working
Papers of the Cornell Phonetics Laboratory 15. 194–270. Ithaca, NY.
Morén, Bruce. 2006. Consonant-vowel interactions in Serbian: Features,
representations and constraint interactions. Lingua 116. 1198–1244.
Morén, Bruce. 2007. Minimalist/substance-free feature theory: Why and how.
Ms., Center for Advanced Study in Theoretical Linguistics, University of
Tromsø.
Oxford, Will. 2012. Patterns of contrast in phonological change: Evidence from
Algonquian vowel systems. Ms., Department of Linguistics, University of
Toronto. To appear in Language.
Pye, Clifton, David Ingram and Helen List. 1987. A comparison of initial
consonant acquisition in English and Quiché. In Keith E. Nelson & Ann
Van Kleeck (eds.), Children's language (vol. 6), 175–190. Hillsdale, NJ:
Erlbaum.
Rice, Keren. 2003. Featural markedness in phonology: Variation. In Lisa Cheng
and Rint Sybesma (eds.),The second Glot International state-of-the-article
book: The latest in linguistics, 387–427. Berlin: Mouton de Gruyter.
32
Rice, Keren. 2007. Markedness in phonology. In Paul de Lacy (ed.), The
Cambridge handbook of phonology, 79–97. Cambridge: CUP.
Rice, Keren & Peter Avery. 1995. Variability in a deterministic model of
language acquisition: A theory of segmental elaboration. In John
Archibald (ed.), Phonological acquisition and phonological theory, 23–
42. Hillsdale, NJ: Lawrence Erlbaum.
Samuels, Bridget D. 2009. The structure of phonological theory. Doctoral
dissertation, Harvard University.
Samuels, Bridget D. 2011. Phonological architecture : A biolinguistic
perspective. Oxford: OUP.
Sandler, Wendy. 1993. Sign language and modularity. Lingua 89. 315–351.
Spahr, Christopher. 2012. Positional neutralization in the Contrastive Hierarchy:
The case of phonological vowel reduction. Ms., Department of
Linguistics, University of Toronto.
Stevens, Kenneth N., Samuel Jay Keyser & Haruko Kawasaki. 1986. Toward a
phonetic and phonological theory of redundant features. In Joseph S.
Perkell and Dennis H. Klatt (eds.), Symposium on invariance and
variability of speech processes, 432–469. Hillsdale, NJ: Lawrence
Erlbaum.
Werker, Janet F., John H. V. Gilbert, Keith Humphrey & Richard C. Tees. 1981.
Developmental aspects of cross-language speech perception. Child
Development 52. 349–355.
33
Werker, Janet F. & Richard C. Tees. 1984. Cross-language speech perception:
Evidence for perceptual reorganization during the first year of life. Infant
Behavior and Development 7. 49–63.
Zhang, Xi. 1996. Vowel systems of the Manchu-Tungus languages of China.
Doctoral dissertation, University of Toronto.
B. Elan Dresher
Department of Linguistics
University of Toronto
Toronto, ON
Canada M5S 3G3
[email protected]
January 2014